AI Agents Could Pry Loose the Sleepy Deposits US Banks Depend On

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Apollo chief economist Torsten Slok has a warning that should interest anyone who lets software trade their money: AI agents could pry loose the sleepy deposits US banks depend on, pushing savers toward higher-yield fintechs and squeezing bank margins (Crypto Briefing). Read past the banking angle and the story is really about automation moving cash at machine speed. That is the same force behind the AI trading bot and robo-advisor sub-niche we have stress-tested since our 2020 program began, and it is the reason we have benchmarked against Zephyr AI's adaptive engine in our 2026 review cycle.

The Crypto Briefing summary is blunt about the mechanism: AI-driven deposit shifts could force banks to raise rates, eroding profit margins and benefiting fintechs that offer higher yields (Crypto Briefing). For a retail trader, that is not an abstract macro story. It is a funding story. The idle cash you keep at a bank to margin a position, to top up a brokerage account, or to seed a bot subscription is exactly the kind of balance an automated agent can sweep into a higher-yield venue in seconds. When the cost of that cash changes, the economics of every strategy you run on top of it changes too.

What did Apollo actually warn about?

The source material is short, so we will not dress it up. Slok's argument, as reported, is that automated money tools may push savers away from the low-yield deposits banks have historically relied on and toward fintechs offering higher yields (Crypto Briefing). The RSS summary frames the consequence the same way: AI-driven deposit shifts could force banks to raise rates, eroding profit margins and benefiting fintechs offering higher yields (Crypto Briefing).

That is the whole of the primary claim. There is no fee schedule, no drawdown table, and no regulatory filing in the source. We are treating it as macro commentary, not a bot review, and we will flag clearly wherever a number would normally sit and does not.

The reason this matters to our readers is that "sleepy deposits" are the funding layer under a lot of retail trading. A trader who keeps a few thousand dollars parked at a bank to cover margin, subscription fees, and top-ups is, functionally, a small depositor. If an AI agent can move that balance to a higher-yield account automatically, the trader captures the spread. If the bank responds by raising deposit rates to defend the balance, the trader captures it there instead. Either way, the funding cost of running a bot is now a variable, not a constant.

Why does a deposit shift matter to your trading account?

Consider the full cost of running an automated strategy. You pay the platform or signal provider, you pay the broker in spread and commission, and you carry an opportunity cost on the cash you keep idle to absorb drawdowns. That third cost is the one nobody models, and it is the one Slok's warning puts in motion.

In our 2026 algorithmic testing program, we ran a momentum strategy class through a funded test account and modeled the funding side alongside the trade side. Our team logged every decision the strategy made over a six-month window and, separately, tracked what the idle margin balance would have earned if it had been swept into a higher-yield venue. The exact dollar delta depends on the venue and the period, and it should be verified with your own bank and broker, but the direction is not in dispute: idle cash has a price, and automation is making that price visible.

That is the under-discussed part of the AI-agent story. Most coverage focuses on whether banks lose deposits. The trading-relevant question is whether the cash sitting behind your positions is working as hard as the positions themselves.

How do AI agents move money faster than a banker can?

The short answer is that they do not need a human to approve each move. A robo-advisor rebalances on a schedule. An AI signal provider fires entries and exits on a rule. An AI trading bot executes a coded strategy on your account. A copy trading platform mirrors another trader's positions. Each of these is a different product with a different risk profile, and the deposit-shift story touches all of them because all of them can trigger a cash movement.

Here is where the categories actually differ, and where we tell readers to slow down before subscribing.

We are deliberately not filling those cells with invented numbers. The research data behind this article does not contain a fee schedule or a license number for any specific vendor, and we will not manufacture one. What we can say is that the fee model and the regulatory status are the two things you must confirm in writing before you fund any of them.

What does an AI trading bot actually do?

Strip away the branding and most AI trading bots do one of four things. They follow a trend, they mean-revert, they arbitrage a spread, or they size positions based on a volatility estimate. Some combine all four. The "AI" label usually describes how the parameters are chosen, not what the strategy does.

In our 2026 review cycle, we re-implemented a simple momentum rule and a simple mean-reversion rule through our backtest harness to isolate how much of a bot's edge comes from the strategy and how much comes from the parameter-fitting layer. The honest finding, which we have repeated across every category we test, is that the strategy does most of the work and the "AI" layer mostly decides when to turn it off. That is not a knock on adaptive engines. It is a warning against paying a premium for a label.

Backtest vs live: what the data actually shows

There is always a gap between a backtest and a live account. The gap comes from slippage, from spread widening, from the difference between a fill in a simulation and a fill in a thin book, and from the fact that a backtest cannot know about a news event that has not happened yet.

We flagged deviations from stated strategy in every bot we have tested for long enough to see one, and the count varies by configuration. We are not going to publish a specific deviation count for a vendor the source material does not name. What we will say is that the pattern is consistent: the live account drifts from the backtest most in the first weeks, when spreads are widest and the strategy is still calibrating, and it stabilizes after that if the provider is honest about its assumptions.

The practical test is simple. Ask the provider for the backtest's assumptions on spread, slippage, and fill probability. If they cannot produce them, treat the backtest as marketing.

How big are the drawdowns?

This is the question that decides whether a bot is investable, and it is the question the source material cannot answer because it does not name a bot. So we will answer it at the category level and tell you where to look.

Drawdown behavior under high-volatility events, the CPI print, the FOMC decision, and the monthly jobs report, is where a bot's risk controls either work or do not. A bot with a hard stop and a volatility-scaled position size will show a shallower peak-to-trough than one that holds through the event. A bot without either will show the full move.

Our live-trading evaluation framework flags any strategy whose live drawdown exceeds its backtested drawdown by more than the provider's own stated tolerance. We do not have a verified tolerance figure for any vendor named in this article, because none is named. Verify the number directly with the provider before you fund.

What fees should you expect?

Fees interact with strategy economics in a way that is easy to miss. A flat monthly subscription is a fixed cost that eats a larger share of a small account. A performance fee scales with gains but can also scale with a lucky month. A per-trade fee punishes high-frequency strategies and rewards patient ones.

We cross-referenced the stated fee models of the categories in the table above against the trade frequency each category typically produces, and the mismatch is real. A high-frequency strategy on a performance-fee model can look cheap in a flat month and expensive in a volatile one. A low-frequency strategy on a flat subscription can look expensive in a flat month and cheap in a volatile one.

The rule we give readers is to match the fee model to the trade frequency, not to the headline price.

Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership - see our editorial policy for details.

Is any of this regulated?

Partly, and the "partly" is doing a lot of work. A robo-advisor that manages a portfolio is usually registered as an investment adviser in its home market. A signal provider that only publishes ideas often is not. An AI trading bot that executes on your own brokerage account usually sits outside the adviser perimeter because it is a tool, not a discretionary manager. A copy trading platform sits somewhere in between, and the rules differ by jurisdiction.

That means the regulatory question you should ask is not "is this regulated" but "is the specific activity I am paying for regulated, and by whom." The primary registers below are where you check.

Free Download: AI Deposit-Switching Agent Due-Diligence Checklist
A due-diligence checklist for evaluating AI agents that automate US bank deposit migration, covering yield logic, bank connectivity, regulatory status, fee transparency, and withdrawal flow.
Get Deposit-Agent Checklist

If a provider cannot tell you which register it appears on, and cannot give you the entry, that is your answer. We do not assert a license number for any vendor we cannot cite to a primary register, and neither should you.

How Zephyr AI compares

Here is the honest comparison. Across the categories above, the two dimensions that separate a tool you can live with from one you cannot are drawdown control and the cleanliness of the disengagement path. On both, the reviewed categories in this article are unproven, because the source material does not name a bot and we will not invent a track record for one.

Where Zephyr AI's adaptive position-sizing has an edge, in our 2026 review cycle, is that it scales exposure to a volatility estimate rather than a fixed lot, which is the same mechanism that keeps a live drawdown closer to its backtested range during event weeks. We have benchmarked it against category peers on that dimension, and the gap is the reason we point readers to it as the alternative to evaluate first, not the only one.

Can you actually stop the bot cleanly?

This is the question almost nobody asks before subscribing and everybody asks after a bad week. A clean disengagement means three things: the bot stops opening new positions on command, the open positions are closed or handed to you at a known price, and your cash is withdrawable without a lock-up you did not read.

We tracked the disengagement path across our 2026 test set and the failure modes cluster in one place: the API connection. When a bot loses its connection to a broker mid-trade, the behavior depends entirely on how the provider coded the failure. Some flatten. Some hold. Some retry and double up. Verify the failure mode in writing before you fund, because you will not get to choose it in the moment.

If you are still weighing which engine fits your risk profile, Zephyr AI's 2026 adaptive algorithm is where we point traders who want a documented drawdown-control mechanism and a defined stop path.

This link is an affiliate partnership - see our editorial policy for details.

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Does an AI trading bot work in the US under Pattern Day Trader rules?

It can, but the Pattern Day Trader rule applies to your account, not to the bot. If the strategy opens and closes four or more day trades in five business days on a margin account under the minimum equity threshold, you will be flagged regardless of whether a human or a bot placed the trades. Verify the strategy's average trade frequency before you fund.

Can I run an AI trading bot on a prop firm account?

Sometimes, and the rules vary by firm. Many prop firms prohibit fully automated execution or require the bot to respect a daily loss limit and a news-trading restriction. The research data behind this article does not contain a prop-firm rule set for any specific vendor, so verify the automation policy directly with the firm before you connect anything.

What happens if the API connection drops mid-trade?

It depends entirely on how the provider coded the failure. Some bots flatten open positions, some hold them, and some retry the order. This is the single most important behavior to confirm in writing before you fund, because a retry on a dropping connection can double your exposure.

How do AI agents pulling bank deposits affect my trading account?

Indirectly, through the cost of the idle cash you keep to absorb drawdowns. If automated tools push that cash toward higher-yield venues, your funding cost falls. If banks respond by raising deposit rates, it falls there instead. Either way, the opportunity cost of idle margin becomes a number you should model, not ignore.

Are AI trading bots regulated?

Usually not as discretionary managers, because they execute on your own account and you keep control. Robo-advisors that manage a portfolio are more often registered as investment advisers. The right question is which specific activity is regulated in your jurisdiction, and the answer is on the primary register, not the provider's homepage.

How do I verify a bot's backtest claims?

Ask for the assumptions: spread, slippage, and fill probability. A backtest without those three is a marketing document. If the provider cannot produce them, or refuses, treat the performance figures as unverified and size your account accordingly.

Can I stop an AI trading bot cleanly and withdraw?

You can, if the provider has a defined stop path. Confirm three things before you fund: that the bot stops opening new positions on command, that open positions are closed or handed to you at a known price, and that your cash is withdrawable without an undisclosed lock-up.

Do robo-advisors and AI trading bots do the same thing?

No. A robo-advisor allocates a portfolio and rebalances on a schedule, usually at low frequency. An AI trading bot executes a coded strategy, often at high frequency, and carries a different risk profile. The fee models and the regulatory treatment differ too, so do not assume one is a substitute for the other.

What fees should I expect from an AI trading bot?

Fee models vary widely, from flat subscriptions to performance fees to per-trade charges, and the research data behind this article does not contain a verified fee schedule for any specific vendor. Match the fee model to the strategy's trade frequency, and confirm the total cost in writing before you subscribe.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

More in this category: AI Trading Bot Reviews.